Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: None
-
Labels:
-
Story Points:1
-
Epic Link:
-
Sprint:DRP S21b
-
Team:Data Release Production
-
Urgent?:No
Description
In doing QA on the gen3 middleware w_2021_16 DC2 processing (of DM-29899), it was noted that the brighter fatter (BF) correction was not being applied (see DM-30351 for a plot and details). As of DM-30351, the appropriate BF kernels and config overrides in obs_lsst have been updated such that the BF correction is now applied by default. The first processing run to include these updates is with w_2021_24 weekly (on DM-30674). The purpose of this ticket is to do a quick demonstration that the BF correction is indeed now being applied. This will be done by a comparison of the "smoking gun" diagnostic plot of the PSF src - model plot for sources that were used in the PSF modeling. Note that this is not a quantification of the detailed (beyond the visual and basic statistics labeled on the plot) performance of the correction or the quality of kernels themselves.
Attachments
Issue Links
Activity
Robert, would you mind having a look at this and let me know if it satisfies the aim of the ticket? The two runs shown here did not technically use the same stack with the sole difference being the config change (but not much in the way of SFM changed between the two weeklies). If you'd still like to see that specific test as more confirmation that the BF correction is now being applied, just let me know and I can run it with the same visit (but noting that a systematic evaluation of the detailed quality of the correction itself would have to be – and I think has been – the topic of a separate ticket!)
I think that looks great. Sorry to be a pest, it just makes me happier to see a differential measurement of a config parameter than config forensics! I'm not worried about the possible tiny SFM changes.
Absolutely no need to apologize...more pestering may help us get in the better habit of doing exactly such important differential and controlled comparisons before changes get merged!
Thanks for the speedy review!
Here is the smoking-gun from the w_2021_20 run having doBrighterFatter=False:
and here is the w_2021_24 version having doBrighterFatter=True:
The disappearance of the offending upwards slope towards the bright end seems to confirm that the correction is indeed now being applied. Note also that the mean at the bright end (S/N>500, mag~<17.3, stats printed at lower left of plots) has decreased by a factor of about two (in the right direction, i.e. closer to zero!) from 0.1992 to 0.0907 percent difference between source and model PSF trace shapes.
I do note that the mean for the S/N>100 sample seems to have a slight offset from zero at about 0.1%), but this is a known issue (as noted in this comment on
DM-6830(unfortunately, it points to a now-defunct HSC Jira ticket for details...)